ResLysEmbed: a ResNet-based framework for succinylated lysine residue prediction using sequence and language model...

Area of Science:

• Biochemistry
• Computational Biology
• Genomics

Background:

• Lysine succinylation is a vital post-translational modification impacting cellular functions and disease development.
• Existing computational tools struggle to accurately predict succinylation sites, hindering research progress.

Purpose of the Study:

• To develop an advanced computational model for precise prediction of lysine succinylation sites.
• To identify optimal protein language models and deep learning architectures for this prediction task.

Main Methods:

• Proposed ResLysEmbed, a novel ResNet-based architecture integrating word and per-residue embeddings from protein language models.
• Compared various protein language models and deep learning architectures, including hybrid models like ConvLysEmbed and InceptLysEmbed.
• Utilized Shapley Additive Explanations (SHAP) for model interpretability.

Main Results:

• ResLysEmbed demonstrated superior performance, achieving high accuracy, MCC, and F1 scores on independent test sets.
• The model outperformed existing methods for succinylation site prediction.
• SHAP analysis provided insights into residue contributions and positional effects on prediction accuracy.

Conclusions:

• ResLysEmbed offers a significant advancement in computational prediction of lysine succinylation.
• The model's interpretability enhances understanding of succinylation mechanisms.
• The developed tool and code are publicly available to facilitate further research.